Applications Manager monitors a Kubernetes cluster by reading the metrics the cluster's Prometheus already collects. On NKP, Prometheus is usually exposed behind a login proxy (traefik-forward-auth with Dex), which a plain integration cannot pass. APM supports this through a Form-based Login authentication mode: it logs in with the same credentials used on the NKP login page, keeps the session, and replays it on every query.

1. Prerequisites

Complete these before adding the integration.

• Prometheus running in the cluster. NKP ships kube-prometheus-stack as a platform application; enable it per cluster or across a workspace. The deployment must include these exporters, which are the data source for APM's Kubernetes monitors:

  • node-exporter — node CPU, memory, disk, and network.
  • kube-state-metrics — Kubernetes object state (pods, deployments, nodes, replicas).
  • kubelet / cAdvisor — pod and container resource usage.

• A discovery label on the scraped targets. APM discovers monitors by a custom label on the metrics. Without it the integration connects but discovers nothing. Add a label with:

  • the same name across clusters, for example apm_filter;
  • the same value for every exporter in a cluster, and a unique value per cluster (for example the cluster name).

  The simplest way is external_labels, which stamps the label on every series from that cluster's Prometheus:

  global:
    external_labels:
      apm_filter: "<unique-cluster-name>"
  

  Per-job control through relabel_configs works too, as long as the label lands on the targets you want monitored.

• Network reachability from the APM server to the cluster's Prometheus endpoint on the required port (443 for the proxy path, 6443 for the API-server path).

2. Choose how APM reaches Prometheus

Pick one path per cluster, based on how Prometheus is exposed.

Form-based Login is the usual choice on NKP because it needs no cluster-side networking change.

3. Add the integration in APM

Go to Settings → Integrations (Add-On Settings) → Prometheus, then fill in:

General specifications

• Display Name: a name for this cluster's integration.
• Poll Interval: collection interval in minutes.
• Mode of data collection: Exposed Prometheus API.

Prometheus Server Details

• Host Name / IP Address: the Prometheus host (for NKP, the cluster endpoint host).
• Prometheus API Port: 443 for the proxy path over HTTPS.
• Endpoint: the path that serves Prometheus, for example /dkp/prometheus.
• Protocol: HTTPS(SSL) for the proxy path.
• Authentication: Form-based Login. Enter the Username and Password used on the NKP login page.
  
  
  

Discovery Specifications

• Monitor Types: select Kubernetes.
• Data Collection Filter Label: the label name you added in the prerequisites, for example apm_filter. This must match the label on the targets.
• Discovery Options: Discover and monitor to create monitors automatically, or Select Monitors to choose specific instances.
• Action on monitor if data not available: how APM treats a monitor that stops returning data.
  
  
  

Save. APM runs a readiness check: it logs in through the proxy, confirms the session, and then discovers monitors using the filter label.

4. Discovery and monitors

• With Discover and monitor, APM creates monitors for the discovered Kubernetes targets and polls them each interval.
• With Select Monitors, use the instances list (searchable) to pick which targets to monitor.
• The proxy session is cached per integration and renewed automatically; if it expires, APM logs in again on the next poll.

5. Monitoring many clusters

Each APM integration points at one Prometheus endpoint, so for multiple clusters you either:

• Add one integration per cluster (the default): isolated and simple, one credential set per cluster.
• Integrate once against a federated or Thanos query endpoint, if NKP Kommander aggregates all clusters. Fewer integrations, but every series must carry the unique per-cluster label so the clusters stay distinguishable.

The per-cluster discovery label from the prerequisites is what keeps clusters apart in either model, so apply it consistently.

6. Troubleshooting